Assaying Out-Of-Distribution Generalization in Transfer Learning

Abstract

Since out-of-distribution generalization is a generally ill-posed problem, various proxy targets (e.g., calibration, adversarial robustness, algorithmic corruptions, invariance across shifts) were studied across different research programs resulting in different recommendations. While sharing the same aspirational goal, these approaches have never been tested under the same experimental conditions on real data. In this paper, we take a unified view of previous work, highlighting message discrepancies that we address empirically, and providing recommendations on how to measure the robustness of a model and how to improve it. To this end, we collect 172 publicly available dataset pairs for training and out-of-distribution evaluation of accuracy, calibration error, adversarial attacks, environment invariance, and synthetic corruptions. We fine-tune over 31k networks, from nine different architectures in the many- and few-shot setting. Our findings confirm that in- and out-of-distribution accuracies tend to increase jointly, but show that their relation is largely dataset-dependent, and in general more nuanced and more complex than posited by previous, smaller scale studies.

References in corpus (17)

• Equality of Opportunity in Supervised Learning
• On Calibration of Modern Neural Networks
• MLP-Mixer: An all-MLP Architecture for Vision
• Zero-Shot Text-to-Image Generation
• AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty
• Underspecification Presents Challenges for Credibility in Modern Machine Learning
• Do ImageNet Classifiers Generalize to ImageNet?
• Fairness in Machine Learning
• Measuring Robustness to Natural Distribution Shifts in Image Classification
• A Large-scale Study of Representation Learning with the Visual Task Adaptation Benchmark
• Fine-Tuning can Distort Pretrained Features and Underperform Out-of-Distribution
• Towards Causal Representation Learning
• Domino: Discovering Systematic Errors with Cross-Modal Embeddings
• Plex: Towards Reliability using Pretrained Large Model Extensions
• The Evolution of Out-of-Distribution Robustness Throughout Fine-Tuning
• MetaShift: A Dataset of Datasets for Evaluating Contextual Distribution Shifts and Training Conflicts
• Dynamic Inference with Neural Interpreters

Cited by in corpus (1)

• Evaluating the Fairness of Discriminative Foundation Models in Computer Vision